An optimized LSTM-based approach applied to early warning and forecasting of ponding in the urban drainage system

Abstract. An optimized LSTM-based approach applied to early warning and forecasting of ponding in the urban drainage system is proposed in this study. This approach can identify locations and process of ponding quickly with relatively high accuracy. The model is constructed with two tandem processes and a multi-task learning mechanism is introduced. The results are compared with those of widely used neural networks (LSTM, CNN) to validate its advantages. Then, the model is revised with available monitoring data in the study area to achieve higher accuracy, and the influence of the number of the monitoring points selected on the performance of the corrected model is also discussed in this paper. Over 15000 designed rainfall events are used for model training, covering a diversity of extreme weather conditions.